This paper reports the extensions of ISO/IEC 25000 (SQuaRE) quality models developed by the Working Group 6 of ISO/IEC JTC1 SC 7 “Software and systems engineering”, to adapt a specific SC7 standard for its use in the field of “Artificial Intelligence” managed by JTC1 SC42. Particular attention is given to the ISO/IEC 25010 as basis of the recent ISO/IEC 25059 for which new quality sub-characteristics have been deemed necessary for the application of the original models to the new technology. The emerging needs are related to the quality of the products to be considered in a complete framework that includes governance, management, implementation processes and evaluation of applications, also quoting some activities on data quality. The article aims to popularize the use of quality models which can mitigate social and ethical risks and increase trust in AI.
This paper briefly describes the evolution of
the quality characteristics of ISO/IEC 25010 [
As mentioned, not all quality subcharacteristics for traditional software and quality in use were found to be satisfactory for the context of AI.
The following table summarizes the new quality sub-characteristics defined by SC42/WG3 for the relative characteristics.
Additional characteristics, such as explainability and safety, are being defined by SC42/WG3 in other complementary standards under development. Also, the evaluation of quality, and related quality measure, is a topic under development in another standard.
Definitions of the new quality subcharacteristics quoted in Table 1 are reported in the clause “Terms and definition” of ISO/IEC DIS 25059 (for the official definitions and more in-deep considerations see Clause 3 and 5). In the following are reported synthesis of the terms: • • • functional correctness: related to the correct results with the needed degree of precision functional adaptability: the AI system can accurately acquire information from data, or the result of previous actions, and is able to use that information in future predictions user controllability: a user can appropriately intervene in an AI system in a timely manner • • • • transparency: if the appropriate information about the AI system is communicated to relevant stakeholders robustness: the AI system can maintain its level of performance under any circumstances intervenability: an operator can intervene in an AI system’s functioning in a timely manner to prevent harm or hazard societal and ethical risk mitigation: the AI system mitigates potential risks to society
For a general vision of concepts and
terminology for AI it is useful to examine the
standard ISO/IEC 22989 “Information technology
- Artificial intelligence - Artificial intelligence
concepts and terminology” [
In the Working Group 2 of ISO/IEC
JTC1/SC42 concerning Data, further activities are
under development. Considering Data an object
complimentary to the Software (years ago data
were considered part of software) it is useful to
mention the importance of data quality for AI, as
appears in the ISO/IEC CD 5259-2 on Data
quality measures [
Since the text of 5259-2 is still in the CDCommittee Draft phase, it is not possible to describe in detail the standard. However, it is useful to mention that also in this standard some new quality characteristics are added, in comparison with other standards taken as a reference. The main additional characteristics to be considered for AI, are those related to groups of data, i.e. datasets, rather than to single data. Various measures of datasets quality, such as dataset representativeness, are also being defined in 5259-2.
The analysis of quality characteristics
included in the data quality model defined in the
standard ISO/IEC 25012 [
The set of documents under development by SC42/WG2 concerning data for AI consists of these ISO/IEC standards: • • • • •
CD 5259-1 Overview, terminology, and examples CD 5259-2 Data quality measures CD 5259-3 Data quality management CD 5259-4 Data quality process framework
AWI 5259-5 Data quality governance
This articulated vision of the topic confirms the importance of data for AI systems.
Quality models are important, but it is also necessary to include some of their characteristics in a higher-level meta-model (like the one called “trustworthiness”) to harmonize multiple factors involved in the production: strategy, governance, management, processes, and products.
The meaning of characteristics of ISO/IEC 25059 suggests not only to apply technical solutions for AI systems, but also to pay attention to the expectations of final users that aim to use products with usability and satisfaction.
Many other AI topics are included in the work of organizations and experts, such as: test, measurement, evaluation, certification, ethics, automated algorithms, laws, regulations, sustainability, human vigilance and governance, accountability, non-discrimination, non-biased data, equity, accessibility, decision making, digital sovereignty, robotics.
The main active organizations for a systemic accepted worldwide view of AI, are, in addition to ISO and others, also the European Commission, the CEN-CLC JTC21, National bodies and other international and national Authorities, Industries, Universities and Associations. Experts and institutions are seeking to build a foundation of trust for AI aiming at improving products and services, mitigating risks, promoting, where useful, data interchange and interoperability of systems. Harmonization of further studies is very desirable. All organizations and experts involved, are called to face the difficult task of building AI systems to improve the well-being of society, environmental protection, and many other contexts.
AI will be a long wave and it will lead to the reformulation of many aspects of traditional software engineering. It is probably essential to carry out research and continuous training at various levels, not only in specific production center but also in the sector of schools and universities.